Find an AI expert: where to look, and how to recognise the right one

The right place depends on one thing only: how clearly your need is already written down. A defined need — “connect our document base to an assistant” — compares well on profiles. A need that is still vague — “we would like to do something with AI” — is better handled by someone whose job is to scope it. The four routes are not mutually exclusive; they simply do not serve at the same moment.

A point of vocabulary that saves time: many companies are looking for an AI consultant for SMEs and type “AI expert”. The two terms refer to the same market, but not to the same service — we come back to this further down. If the choice between a firm and a freelancer has not yet been made, it is covered separately: AI agency or AI freelancer, which to choose.

There is no official list of AI experts in France: no state register, no mandatory qualification, no professional body. Anyone who publishes a ranking of the “best AI experts” has therefore built it on their own criteria, and it is up to them to publish those criteria. What we can say on our side is verifiable: this directory publishes 399 profiles of French-speaking AI providers across 16 countries, 323 of them in France — a count recalculated every time the site goes live, not an ageing snapshot.

The only official qualification in the Claude ecosystem is Anthropic’s programme. Its first tier, Registered, requires a commitment to train ten people; the Select tier, the first to list a provider in Anthropic’s directory, requires ten certified people and two Claude customers in production over twelve months (thresholds noted on 24 August 2026 from Anthropic’s announcement of 3 June 2026). In other words: a very good freelancer will never appear there, and their absence says nothing about their work. On our side, we award a single title, the Claude Partners Certified badge, earned through a free test; it is shown on the profile but does not change its rank. These ranking rules are published, and the site’s only two paid placements are named there.

This directory sets a grid of €300 to €700 excl. VAT per day for a freelance AI expert, depending on experience — a market benchmark set by the platform, not a measurement of the rates actually charged. The rates that providers have given us themselves, listed on the barometer, often fall outside this range, both above and below. For an agency, we publish no order of magnitude: we do not measure one.

This grid is never the rate of a named person: a profile only shows a rate if the provider gave it themselves. If you are looking instead for the cost of an AI agent — a piece of software, not a person —, it is a project budget you need to estimate, not a day rate: how much an AI project costs does that calculation. The barometer’s method and its breakdown by seniority are published on the day-rate barometer. On the salary of an employed AI expert — another question Google associates with this one — we measure nothing, and we say so on the Claude AI expert page.

Three signals can be read on a profile, without a meeting, and rule out most of the wrong leads. They do not replace the interview; they save you from sitting through five for nothing.

1. A named deliverable, not a declared skill. “Expertise in generative AI” tells you nothing. “Tender-response assistant deployed at a 40-person engineering consultancy” tells you the scope, the client size and the trade. A profile that only lists technologies describes keeping up to date, not a practice.

2. Consistency between what is claimed and what is described. A profile that ticks eleven areas of expertise and details a single project mainly tells you it does not want to miss any search. Two or three areas genuinely documented are worth more than the full list.

3. How long they have really been doing AI engagements, not how old the company is. Many firms have existed for ten years and have been doing AI for six months. The two dates are different, and it is the second that matters to you.

Then comes the interview, where three questions are enough to decide — what has been put into production and for whom, how the model connects to your systems, and where your data goes. They are developed on the page dedicated to the Claude AI expert, with the answer you should expect.

None of these three titles is protected: they describe a role in your project, not a level. AI expert is the broadest term, and it is the one companies type; it presumes nothing. The AI consultant scopes, arbitrates and trains: you call on them when the question is “where do we start”. The integrator builds and puts into production: you call on them when the answer is already known.

Many providers do both, and that is fine — but paying for three weeks of scoping when the need was written down from the start remains the most frequent expense, and the most avoidable. To see who does what, the directory arranges profiles by area of expertise and by sector: browse the 399 profiles.